1、BioLINK Talks,BioLINK,Detroit, June 24 (Edinburgh July 11),Linking Literature, Information and Knowledge for Biology,Corpora and Corpus design (2)NER and Term Normalisation (3)Annotation and Zoning (2)Relation Extraction (2)Other,Corpus Design for Biomedical Natural Language Processing,K. Bretonnel
2、Cohen et al (U of Colorado),Main Question: why are some (bio-)corpora more used than others? What makes them attractive?,Crucial points:,Take home message: if you want people to use your corpus, use XML, publish annotation guidelines, publicise corpus with dedicated papers, use it for competitions,f
3、ormat: XMLcode several layers of informationpublicity: write specific papers about corpus, publicise its availability,Corpora and corpus design,MedTag: a collection of biomedical annotations,L. Smith et al. (National Center for Biotechnology Information, Bethesda, Maryland),Main Point: MedTag is a d
4、atabase that combines three corpora:,Take home message: integrated data, more accessible, you should try it.,Corpora and corpus design,MedPost (modified to include 1000 extra sentences)ABGeneGENETAG (modified to reflect new defs of genes and prots),The data is available in flat files + software to f
5、acilitate loading data into SQL database,MedPost,6700 sentencesannotated for POS and gerund argumentsPOS tagger trained on it (97.4% accuracy),GENETAG,15000 sentences currently released tagged for gene/protein identificationused in Biocreative,ABGene,over 4000 sentencesannotated for gene/protein nam
6、esNER tagger trained on it (lower 70s),Corpora and corpus design,GOOD,BAD,Recommended Usestraining and evaluatingPOS taggerstraining and evaluatingNER taggersdeveloping and evaluatinga chunker (for PubMedphrase indexing)analysis of grammatical usage in medical textfeature extraction for MLentity ann
7、otation guidelines,tokenisation! (white spaces were deleted),Corpora and corpus design,NER and TN,Weakly Supervised Learning Methods for Improving the Quality of Gene Name Normalization,Ben Wellner (MITRE),1. presenting method of improving quality of training data from BioCreative task1b. Systems pe
8、rformance on improved data is better than on original data2. weakly supervised methods can be successfully appliedfor re-labeling noisy training data,Main points,(next week),NER and TN,Unsupervised gene/protein normalization using automatically extracted dictionaries,A. Cohen (Oregon Health & Scienc
9、e U., Portland, Oregon),Main point: dictionary-based gene and protein NER and normalisation system; no supervised training; no human intervention.,what curated databases are the best collections of names?are simple rules sufficient for generating ortographic variants?can common English words be used
10、 to decrease false positives?what is the normalization performance of a dictionary-based approach?,Results: near state-of-the-art; saving on annotation,METHOD,1. Building the dictionary,2. Generating orthographic variants,3. Separating common English words,4. Screening out most common English words,
11、5. Searching the text,6. Disambiguation,Automatically extracted from 5 databases: official symbol, Unique identifiers, name, symbol, synonym, alias fields,Set of 7 simple rules applied iteratively,Dictionary split in two parts: confusion and main dictionary,Note: 5% ambiguous intra-species; 85% acro
12、ss species. Exploit non-ambiguous synonyms; exploit context,NER and TN,NER and TN,A machine learning approach to acronym generation,Tsuruoka et al (Tokyo (Tsujii group), Japan and Salford, UK),Task: system generates possible acronyms from a given expanded form,Method: ML approach (MaxEnt Markov Mode
13、l),Main point: acronym generation as sequence tagging problem,Experiments:- 1901 definition/acronym pairs- several ranked options as output- 75.4% coverage when including top 5 candidates- baseline: take first letters and capitalise them,Classes (tags)1. SKIP (generator skips the letter) 2. UPPER (g
14、enerator upper-cases letter) 3. LOWER (generator lower-cases letter) 4. SPACE (generator converts letter into space) 5. HYPHEN (generator converts letter into hyphen),Features- letter unigram - letter bigram - letter trigam - action history (preceding action) - orthographic (uppercase or not) - leng
15、th (#words in definition) - letter sequence - distance (between target letter and beginning/tail of word),NER and TN,Searching for High-Utility Text in the Biomedical Lit.,Shatkay et al. (Queens,Ontario and NYU and NCBI,Maryland),(Main idea + annotation guidelines),High Utility Regions = regions in
16、the text that we identify as focusing on scientific findings, stated with a high confidence, and preferably supported by experimental evidence.,Task: identify text regions that are rich in scientific content, and retrieve docs that have many such regions,Annotation/Zoning,assertion = sentence or fra
17、gmentFocus = type of information conveyed by assertion- scientific- generic- methodologyPolarity of assertion (positive/negative)Certainty- complete uncertainty (0)- complete certainty (3)Evidence = whether assertion is supported by exp evidence- E0 = lack of evidence- E1 = evidence exists but not r
18、eported (“it was shown”)- E2 = evidence not given directly but reference provided- E3 = evidence providedDirection/Trend = whether assertion reports increase/decreasein specific phenomenon,K=.83,K=.81,K=.70,K=.73,K=.81,Annotation/Zoning,Automatic Highlighting of Bioscience Literature,Annotation/Zoni
19、ng,H. Wang et al (CS Department, University of Iowa - M. Light group),Task: automatic highlighting of relevant passages,Approach: IR task- sentence is passage unit- each sentence treated as document- user provides a query- query box for keywords- example passage highlighting- system ranks sentences
20、as to relevance to query(* query expansion system is web-based),Annotation/Zoning,- Corpus: 13 journal articles each highlighted by a bio graduate student before the request for annotation,- Queries: constructed in retrospect. The annotators createdthe queries for the articles they had selected. The
21、 first highlighted region also used as query,- Processing: tokenisation (LingPipe), indexing (Zettair), rankingof retrieved sentences (Zettair),- Query Expansion: definitions were used. Google “define”for each word (excluding stopwords). Over 80% of query words had Google defs.,poor results first hi
22、ghlighted passage works better than keywordsGoogle expansion helps,Using biomedical literature mining to consolidate the set of known human PPIs,Rel Extr,A. Ramani et al (U of Texas at Austin - Bunescu/Mooney group),Task: construct a database of known human PPIs by:- combining and linking interactio
23、ns from existing DBs- mine additional interactions from 750000 Medline abs,Results: - quality of automatically extracted interactions comparableto that of those extracted manually- overall network of 31609 interactions between 7748 prots,1. Identify proteins in text: CRF tagger2. Filter out less con
24、fident entities3. Try to detect which pairs of remaining ones are interactions,- use co-citation analysis - train model on existing set,Trained model: a sentence containing 2 protein names is classified as correct/wrong. If a sentence has n prots (n 2), the sentence is replicated n times- ELCS = Ext
25、raction w Longest Common Subsequences (learned rules) - ERK = Extraction using a Relation Kernel,Rel Extr,IntEx: A syntactic role driven PPI extractor for biomedical text,Rel Extr,S. Ahmed et al (Arizona State University),Task: detect PPIs by reducing complex sentences to simple clauses and then exp
26、loiting syntactic relations,- pronoun resolution (third person and reflexives; simple heuristics) - entity tagging (dictionary lookup + heuristics) - parsing (Link Grammar, dependency based, CMU?) - complex sentence splitting (verb-based approach to extract simple clauses) - interaction extraction (from simple clauses exploiting syntactic roles),
